High-frequency stock trading and prediction using feedforward neural networks
Date of Issue2018
School of Electrical and Electronic Engineering
In this project, I use feed-forward neural networks to predict the high frequency stock price. I study some of the published methods used to predict the stock price and choose two of them to implement and reproduce the experiment results with the same data source. The first published method is to use recent days’ price such as open price, highest price, lowest price, close price etc as input features to predict the next day’s price. The second published method is to use recent minutes’ change of the price, trend and standard deviations of prices as input features to predict the next minute’s change. I explore the approaches of improving the prediction performance for each method. For the first method, no approach is found to improve the prediction accuracy obviously. However, for the second method, I propose two approaches to improve the prediction performance for the next hour’s average price trend. The first approach improves the directional accuracy from 69.50% to 82.23% by adding cut-off points at the prediction output to filter out high quality predictions. The second approach improves the directional accuracy from 69.50% to 75.14% by combining three networks’ outputs to making the prediction.
DRNTU::Engineering::Electrical and electronic engineering
Final Year Project (FYP)
Nanyang Technological University